Coordinated CESM/CanESM Large Ensembles for the CanSISE Community Paul Kushner Department of Physics University of Toronto.

Coordinated CESM/CanESM Large Ensembles for the CanSISE Community

Paul KushnerDepartment of PhysicsUniversity of Toronto

Purpose of this Discussion• CanSISE Theme B and Deliverable 3:

Attribution of Climate/Cryospheric Events (ACE/ACRE)– Quantify probability of extremes in presence or absence of

anthropogenic influence (FAR).– E.g. what is the fraction of risk attributable to human influence of

the 2007 vs. the 2012 sea ice minimum?• Create large initial condition ensembles of historical and

projected forcing simulations.– Coordinated between CanESM (CCCma) and NCAR CESM1

(Toronto).– Carry out attribution runs with selected radiative forcings.

• Ensemble size large enough to simultaneously estimate – forced response and – probability distribution associated with internal variability.

Purpose of this Discussion• But we can use such runs for many other

purposes than ACE/ACRE!– Let’s look on them as a community resource.– Consider the design and diagnostic aims.

• To jog ideas I will review some applications of the large ensemble approach, then talk about our proposed experiments.

Distribution of Global Temperature

Trends in CESM Large Ensemble

(Kay et al. 2014, submitted to BAMS)

Attributing changes in extreme temperatures to

model uncertainty versus internal variability

(Fischer et al. 2014, Nature Climate Change)

Communicating uncertainty in

regional trends arising from internal

variability

(Deser et al. 2012, Nature Climate

Change)

Analyzing cryospheric trends

in context of internal variability and observational uncertainty.

(Mudryk et al. 2013, Climate Dynamics)

Several Large Ensemble ProjectsModel Time Period Forcing N Projects

CCSM3 2005-2065 A1B Projection

40 Deser et al., Teng/Branstator

CCSM4 1955-2010 Historical 40 Mudryk et al.

CESM1-CAM5 1950-2100 Historical/rcp8.5

21 Fischer et al.

CESM1-CAM5 1920-2080 Historical/rcp8.5

30 (NCAR) 1 + 6 (UofT)

Kay et al.

CanESM and NCAR CESM (U of T)

1960-2020

Historical-ALL FORCE

50 CanSISE Theme BHistorical-NAT

In red are shown runs to be done this year.

Design DiscussionModel version • CanESM – which version?

• CESM1 – CAM5: 2 degree atmosphere, 1 degree oceanForcing • Generally, use each model’s forcings.

• But CESM has specially developed stratospheric ozone from WACCM with realistic Antarctic Ozone Hole.• Should this be used for CanESM?

pictrl spin-up • Does this exist for CanESM version used?• To be done in 2014 for CESM1 2 degree on SciNET (Compute

Canada allocation)Initialization and generation of realizations

1850-1950, branch at 1950, analysis 1960-2020• Does this exist for CanESM version used?• To be done in 2014 for CESM1 on SciNet• Standard perturbation methods to generate realizations.

Data saved • Diagnostic set developed for CESM project.• Extensive, but a good starting point.• NetCDF data compression reduces data by factor of 2-4.• Each 3-D variable saved monthly requires ~15-20GB for 2

degree model resolution on 30 levels. • What output should be saved and where should it be saved?

Design DiscussionModel version • CanESM – which version?

• CESM1 – CAM5: 2 degree atmosphere, 1 degree oceanForcing • Generally, use each model’s forcings.

• But CESM has specially developed stratospheric ozone from WACCM with realistic Antarctic Ozone Hole.• Should this be used for CanESM?

pictrl spin-up • Does this exist for CanESM version used?• To be done in 2014 for CESM1 2 degree on SciNET (Compute

Canada allocation)Initialization and generation of realizations

1850-1950, branch at 1950, analysis 1960-2020• Does this exist for CanESM version used?• To be done in 2014 for CESM1 on SciNet• Standard perturbation methods to generate realizations.

Data saved • Diagnostic set developed for CESM project.• Extensive, but a good starting point.• NetCDF data compression reduces data by factor of 2-4.• Each 3-D variable saved monthly requires ~15-20GB for 2

degree model resolution on 30 levels. • What output should be saved and where should it be saved?

Design DiscussionModel version • CanESM – which version?

• CESM1 – CAM5: 2 degree atmosphere, 1 degree oceanForcing • Generally, use each model’s forcings.

• But CESM has specially developed stratospheric ozone from WACCM with realistic Antarctic Ozone Hole.• Should this be used for CanESM?

pictrl spin-up (multi century)

• Does this exist for CanESM version used?• To be done in 2014 for CESM1 2 degree on SciNET (Compute

Canada allocation)Initialization and generation of realizations

1850-1950, branch at 1950, analysis 1960-2020• Does this exist for CanESM version used?• To be done in 2014 for CESM1 on SciNet• Standard perturbation methods to generate realizations.

Data saved • Diagnostic set developed for CESM project.• Extensive, but a good starting point.• NetCDF data compression reduces data by factor of 2-4.• Each 3-D variable saved monthly requires ~15-20GB for 2

degree model resolution on 30 levels. • What output should be saved and where should it be saved?

Design DiscussionModel version • CanESM – which version?

• CESM1 – CAM5: 2 degree atmosphere, 1 degree oceanForcing • Generally, use each model’s forcings.

• But CESM has specially developed stratospheric ozone from WACCM with realistic Antarctic Ozone Hole.• Should this be used for CanESM?

pictrl spin-up (multi century)

• Does this exist for CanESM version used?• To be done in 2014 for CESM1 2 degree on SciNET (Compute

Canada allocation)Initialization and generation of realizations

1850-1950, branch at 1950, analysis 1960-2020• Does this exist for CanESM version used?• To be done in 2014 for CESM1 on SciNet• Standard perturbation methods to generate realizations.

Data saved • Diagnostic set developed for CESM project.• Extensive, but a good starting point.• NetCDF data compression reduces data by factor of 2-4.• Each 3-D variable saved monthly requires ~15-20GB for 2

degree model resolution on 30 levels. • What output should be saved and where should it be saved?

Design DiscussionModel version • CanESM – which version?

• CESM1 – CAM5: 2 degree atmosphere, 1 degree oceanForcing • Generally, use each model’s forcings.

• But CESM has specially developed stratospheric ozone from WACCM with realistic Antarctic Ozone Hole.• Should this be used for CanESM?

pictrl spin-up (multi century)

• Does this exist for CanESM version used?• To be done in 2014 for CESM1 2 degree on SciNET (Compute

Canada allocation)Initialization and generation of realizations

1850-1950, branch at 1950, analysis 1960-2020• Does this exist for CanESM version used?• To be done in 2014 for CESM1 on SciNet• Standard perturbation methods to generate realizations.

Data saved • Diagnostic set developed for CESM project.• Extensive, but a good starting point.• NetCDF data compression reduces data by factor of 2-4.• Each 3-D variable saved monthly requires ~15-20GB for 2

degree model resolution on 30 levels. • What output should be saved and where should it be saved?

Conclusion

• Large initial condition ensembles involve work to coordinate, significant computation, data resources.

• They can add value to traditional CMIP sets, but this requires good planning/coordination. Let’s make sure to coordinate well.

• Tentative timeline:– June 2014: finalize design– August 2014: pictrl done– December 2014: 5-10 realizations done, initial comparisons.– April 2015: full sets done.

Discussion